MATLAB Implementation of Blind Deconvolution Algorithm

Blind deconvolution is a technique for recovering blurred images when the point spread function (PSF) is unknown. The algorithm proposed by Israeli researchers employs an iterative approach to simultaneously estimate the original sharp image and the blur kernel (PSF), ultimately achieving image deblurring.

The core methodology of this algorithm can be divided into three stages: initialization, alternating optimization, and result refinement. First, a rough PSF estimate is initialized based on gradient information or spectral characteristics of the blurred image. The algorithm then proceeds to the alternating optimization phase - fixing the current PSF estimate and using non-blind deconvolution techniques to restore the image, followed by fixing the restored image to update the PSF estimate. This alternating process progressively improves the accuracy of both components through constrained optimization (such as sparsity constraints or energy minimization).

During iterations, regularization terms are introduced to prevent noise amplification, with common techniques including total variation regularization or L1/L2 norm constraints. The algorithm's termination condition is typically set when the PSF estimate variation falls below a threshold or when the maximum iteration count is reached. The final output sharp image undergoes post-processing (such as contrast enhancement) to improve visual quality.

The advantage of this method lies in its strong adaptability, capable of handling various degradation scenarios like motion blur and defocus blur without requiring pre-calibrated device parameters. In practical applications, attention should be paid to noise sensitivity issues, and it's recommended to incorporate a noise estimation module.